Datasets:

Adonis3039
/

ELGAR-Bench

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
schema_version: int64
task_id: string
split: string
gold_facts: list<item: struct<fact_id: string, claim: string, value: string, unit: string, entity: string, time: (... 78 chars omitted)
  child 0, item: struct<fact_id: string, claim: string, value: string, unit: string, entity: string, time: string, me (... 66 chars omitted)
      child 0, fact_id: string
      child 1, claim: string
      child 2, value: string
      child 3, unit: string
      child 4, entity: string
      child 5, time: string
      child 6, metric: string
      child 7, evidence_ids: list<item: string>
          child 0, item: string
      child 8, raw_value: string
gold_evidence: list<item: struct<evidence_id: string, source_file: string, page: int64, text: string, component: st (... 368 chars omitted)
  child 0, item: struct<evidence_id: string, source_file: string, page: int64, text: string, component: struct<id: st (... 356 chars omitted)
      child 0, evidence_id: string
      child 1, source_file: string
      child 2, page: int64
      child 3, text: string
      child 4, component: struct<id: string, label: string, layer: string, type: string>
          child 0, id: string
          child 1, label: string
          child 2, layer: string
          child 3, type: string
      child 5, page_or_pointer: string
      child 6, relation: struct<source: string, target: string, label: string, style: string>
          child 0, source: string
          child 1, target: string
          child 2, label: st
...
ct_ids: list<item: string>
              child 0, item: string
          child 5, claim_ids: list<item: string>
              child 0, item: string
          child 6, source_refs: list<item: struct<claim_id: string, source_file: string, page_or_pointer: string>>
              child 0, item: struct<claim_id: string, source_file: string, page_or_pointer: string>
                  child 0, claim_id: string
                  child 1, source_file: string
                  child 2, page_or_pointer: string
          child 7, claim_id: string
          child 8, source_file: string
          child 9, page_or_pointer: string
required_outputs: struct<table: bool, chart: bool, diagram: bool, source_image: bool>
  child 0, table: bool
  child 1, chart: bool
  child 2, diagram: bool
  child 3, source_image: bool
failures: list<item: null>
  child 0, item: null
version: string
passed: bool
failure_count: int64
evidence_count: int64
task_count: int64
match_method_counts: struct<exact_adjacent_cross_page_method_span: int64, exact_djvu_ocr_page_words: int64, exact_docx_pa (... 212 chars omitted)
  child 0, exact_adjacent_cross_page_method_span: int64
  child 1, exact_djvu_ocr_page_words: int64
  child 2, exact_docx_paragraph: int64
  child 3, exact_html_xpath: int64
  child 4, exact_json_pointer: int64
  child 5, exact_normalized_pdf_page: int64
  child 6, exact_pdf_page_text: int64
  child 7, exact_pptx_slide_paragraph: int64
  child 8, exact_xlsx_cell: int64
  child 9, exact_xml_xpath: int64
to
{'schema_version': Value('int64'), 'version': Value('string'), 'passed': Value('bool'), 'failure_count': Value('int64'), 'failures': List(Value('null')), 'task_count': Value('int64'), 'evidence_count': Value('int64'), 'match_method_counts': {'exact_adjacent_cross_page_method_span': Value('int64'), 'exact_djvu_ocr_page_words': Value('int64'), 'exact_docx_paragraph': Value('int64'), 'exact_html_xpath': Value('int64'), 'exact_json_pointer': Value('int64'), 'exact_normalized_pdf_page': Value('int64'), 'exact_pdf_page_text': Value('int64'), 'exact_pptx_slide_paragraph': Value('int64'), 'exact_xlsx_cell': Value('int64'), 'exact_xml_xpath': Value('int64')}}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              schema_version: int64
              task_id: string
              split: string
              gold_facts: list<item: struct<fact_id: string, claim: string, value: string, unit: string, entity: string, time: (... 78 chars omitted)
                child 0, item: struct<fact_id: string, claim: string, value: string, unit: string, entity: string, time: string, me (... 66 chars omitted)
                    child 0, fact_id: string
                    child 1, claim: string
                    child 2, value: string
                    child 3, unit: string
                    child 4, entity: string
                    child 5, time: string
                    child 6, metric: string
                    child 7, evidence_ids: list<item: string>
                        child 0, item: string
                    child 8, raw_value: string
              gold_evidence: list<item: struct<evidence_id: string, source_file: string, page: int64, text: string, component: st (... 368 chars omitted)
                child 0, item: struct<evidence_id: string, source_file: string, page: int64, text: string, component: struct<id: st (... 356 chars omitted)
                    child 0, evidence_id: string
                    child 1, source_file: string
                    child 2, page: int64
                    child 3, text: string
                    child 4, component: struct<id: string, label: string, layer: string, type: string>
                        child 0, id: string
                        child 1, label: string
                        child 2, layer: string
                        child 3, type: string
                    child 5, page_or_pointer: string
                    child 6, relation: struct<source: string, target: string, label: string, style: string>
                        child 0, source: string
                        child 1, target: string
                        child 2, label: st
              ...
              ct_ids: list<item: string>
                            child 0, item: string
                        child 5, claim_ids: list<item: string>
                            child 0, item: string
                        child 6, source_refs: list<item: struct<claim_id: string, source_file: string, page_or_pointer: string>>
                            child 0, item: struct<claim_id: string, source_file: string, page_or_pointer: string>
                                child 0, claim_id: string
                                child 1, source_file: string
                                child 2, page_or_pointer: string
                        child 7, claim_id: string
                        child 8, source_file: string
                        child 9, page_or_pointer: string
              required_outputs: struct<table: bool, chart: bool, diagram: bool, source_image: bool>
                child 0, table: bool
                child 1, chart: bool
                child 2, diagram: bool
                child 3, source_image: bool
              failures: list<item: null>
                child 0, item: null
              version: string
              passed: bool
              failure_count: int64
              evidence_count: int64
              task_count: int64
              match_method_counts: struct<exact_adjacent_cross_page_method_span: int64, exact_djvu_ocr_page_words: int64, exact_docx_pa (... 212 chars omitted)
                child 0, exact_adjacent_cross_page_method_span: int64
                child 1, exact_djvu_ocr_page_words: int64
                child 2, exact_docx_paragraph: int64
                child 3, exact_html_xpath: int64
                child 4, exact_json_pointer: int64
                child 5, exact_normalized_pdf_page: int64
                child 6, exact_pdf_page_text: int64
                child 7, exact_pptx_slide_paragraph: int64
                child 8, exact_xlsx_cell: int64
                child 9, exact_xml_xpath: int64
              to
              {'schema_version': Value('int64'), 'version': Value('string'), 'passed': Value('bool'), 'failure_count': Value('int64'), 'failures': List(Value('null')), 'task_count': Value('int64'), 'evidence_count': Value('int64'), 'match_method_counts': {'exact_adjacent_cross_page_method_span': Value('int64'), 'exact_djvu_ocr_page_words': Value('int64'), 'exact_docx_paragraph': Value('int64'), 'exact_html_xpath': Value('int64'), 'exact_json_pointer': Value('int64'), 'exact_normalized_pdf_page': Value('int64'), 'exact_pdf_page_text': Value('int64'), 'exact_pptx_slide_paragraph': Value('int64'), 'exact_xlsx_cell': Value('int64'), 'exact_xml_xpath': Value('int64')}}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

ELGAR-Bench: peer-review development subset

This Hugging Face package is a partial submission-stage release for EviWeave: Provenance-Guided Visual Planning for Verifiable Multimodal Reports. It contains 235 development tasks, their development gold labels, the referenced source documents, and per-document provenance/licence metadata.

Scope

The full benchmark contains 881 tasks over 1,073 documents in 451 source clusters: 250 development, 250 frozen local holdout, and 381 challenge-OOD. This package releases only 235 development tasks whose source tier permits redistribution or is public. Fifteen FinanceBench CC-BY-NC development tasks are excluded. The frozen local holdout, challenge-OOD tasks, all corresponding gold labels, and human-review packets are withheld during peer review.

Files

public/tasks.dev.jsonl: public task specifications.
private/gold.dev.jsonl: gold labels for the released development tasks only; private/ is retained solely for compatibility with the scorer.
public/documents/: source files referenced by the released tasks.
public/document_manifest.jsonl: document path, checksum, source dataset, publisher, language, and transformation metadata. Licence and access-tier fields are retained in each row of public/tasks.dev.jsonl.
metadata/full_benchmark/: aggregate machine-audit and calibration records.
metadata/human_review/: non-identifying review summaries only.
metadata/frozen_local_holdout_protocol.json: hashes and run-lock metadata; no held-out inputs or labels are included.
release_manifest.json: exact inclusion/exclusion counts and selection rule.

Access and licence

There is no single licence covering every source document. The source dataset and publisher are recorded in public/document_manifest.jsonl; the applicable licence and access tier are recorded per task in public/tasks.dev.jsonl and summarized in release_manifest.json. Do not apply a blanket dataset licence to third-party files. The package is suitable for a private or gated reviewer repository until the authors complete a final source-URL and redistribution audit.

Evaluation

Use the separate EviWeave code package:

python scripts/run_elgar_bench.py \
  --dataset-dir /path/to/ELGAR-Bench \
  --split dev \
  --output-dir /path/to/predictions

The full 250-task local holdout was frozen by input/model/template hashes and a runtime/code snapshot, then run once. It is a local holdout, not an external secret-test server. Its labels are not present here and were not serialized into model-facing commands.

Downloads last month: 38

Total file size:

3.69 MB